The 180-Degree Turn
At the end of 2025, the compute market was still worried about oversupply.
By early 2026, the winds had completely shifted. Compute became scarce, prices rose. Kimi's API revenue during just a few days of the Spring Festival equaled its entire previous year.
The reason isn't complicated: how AI is used has changed. From chat to agents, compute consumption increased 100-fold in just two to three months. Data shows that in agent scenarios, per-user token consumption is 10 to 50 times that of regular chat, with some scenarios seeing increases of up to 370x.
Cloud compute is being utilized quite fully—tech giants scrambling for GPUs, startups waiting in line, resources rarely going to waste.
But edge compute is a different story.
The Edge Dilemma
AI PCs, edge inference servers, various AI accelerators—these devices are developing rapidly, and prices are falling fast. NVIDIA's DGX Spark costs a little over 30,000 yuan, while AMD's Ryzen AI Max+ 395 is even under 20,000. The compute is substantial, capable of inferencing fairly large models.
But are they being used well?
The answer I see is: no.
The problem lies in Total Cost of Ownership (TCO). The devices themselves are cheaper, but the proportion of other cost items has been violently magnified.
Operations and Maintenance Costs
A server costing 200,000 yuan that fails twice a year, costing 1,000 yuan each time for technical support—that's 2,000 yuan annually. That's 1% of the device price, negligible.
Switch to a 20,000 yuan edge device. The failure rate won't be lower—it might even be higher, because edge environments are more complex. Still 2,000 yuan in O&M costs.
But now that's 10% of the price. From 1% to 10%—a 10x magnification.
The reality is worse. On-site maintenance for edge devices can cost 500 yuan per visit per device, and with scattered deployments, inspection efficiency is low.
Cold Start Costs
A device fresh from the factory is just a piece of metal.
How do you turn it into real AI compute? How do you extract its performance?
This is harder than imagined. Hardware evolves daily, inference engines vary widely, models change too fast, heterogeneous architectures cause headaches. ARM, x86, RISC-V coexist; firmware versions are fragmented. One company with 2,000 edge devices saw troubleshooting time increase 3x due to inconsistent firmware versions.
To use these devices well, you need someone who understands hardware, inference engines, models, and applications. A monthly salary of 20,000 yuan for such a person isn't unreasonable.
But the device itself only costs 20,000 yuan.
That's equivalent to paying a 100% premium just to use one device. That doesn't make business sense.
Continuous Usage Costs
Getting the device running isn't the end—it's the beginning.
In 2026, major model providers are updating at dizzying speeds. Every month or two brings a major new model release. Google Gemini iterated from 2.5 to 3 Flash within a year, Alibaba's Qwen3 series launched intensively, DeepSeek and Kimi appeared in succession.
If your edge device is still running a year-old model, its value might be only 10% of what it was.
Keeping up with this pace means continuous upgrading, adaptation, testing, and deployment. These costs are calculated per headcount—they don't get discounted just because the devices are cheap.
Do People Really Want to Be Involved?
Traditional software's solution approach is: design a system for people to operate.
This approach has an implicit assumption: people are willing to participate.
But in the edge AI scenario, does this assumption hold?
Think about electricity. People use it but don't participate in its production and management. Electricity is infrastructure—people just use it. No one feels the need to "participate in the production of electricity."
Edge AI devices should be the same.
They are infrastructure, not tools. The ideal state is devices that self-run, self-repair, and self-upgrade, never failing.
If people don't have to participate, all the better.
This means a shift in design thinking: let devices manage themselves, keep humans outside the loop, software isn't a tool for people but the "nervous system" of the device. The goal isn't reducing human burden—it's eliminating human participation.
How to Achieve This?
I think there are several trends we can leverage.
First, AI Agent capabilities are rapidly strengthening. If your product doesn't let Agents participate in core processes, it's out of step with the times. Agents can complete O&M diagnosis, fault repair, performance tuning, and model upgrades without human intervention—automating what that "20,000 yuan/month person" does. This isn't future tense; 2026 Agent capabilities are already sufficient.
Another is open-source community collaboration. Edge AI faces a four-dimensional optimization problem: hardware devices × inference engines × models × applications. Every dimension is changing rapidly; no single company can keep up alone. But with Agent proliferation, collaboration based on code repositories becomes easier. Products need to think about how to get more people and Agents involved to jointly address growing complexity.
Finally, internet infrastructure. Edge devices are characterized by fragmentation—AI PCs and edge servers distributed across offices, factories, stores, and homes, unlike data centers with thousands of machines in one room. To extract value from these scattered devices, you must use the internet. Remote monitoring, OTA upgrades, cloud collaboration—these aren't optional.
Final Thoughts
The TCO problem of edge AI is essentially a management philosophy problem.
Continuing to design products with a "human-centered" approach means O&M costs will keep amplifying, cold start costs will remain high, and continuous usage costs will keep devouring value.
The solution is to let AI participate. Use Agents to automate O&M, use community to share complexity, use internet to connect scattered devices.
This is what we're doing.